MaskID: An effective deep-learning-based algorithm for dense rebar counting.

Area of Science:

• Computer Vision
• Deep Learning
• Robotics Automation

Background:

• Dense rebar counting is crucial for automation but challenging due to overlapping objects, varied sizes, and poor lighting.
• Traditional manual and machine vision methods lack efficiency and accuracy in complex environments like rebar yards.

Purpose of the Study:

• To develop an accurate and efficient deep learning-based method for dense rebar counting.
• To address the limitations of existing methods in handling complex environmental conditions and object occlusions.

Main Methods:

• An improved Mask R-CNN model was developed for detailed segmentation of large-scale rebar images via splitting and rescaling.
• Intelligent denoising strategies were applied to enhance segmentation accuracy by addressing object overlap, aspect ratios, distribution, and environmental noise.

Main Results:

• The proposed MaskID method achieved an F1-score exceeding 0.99 across multiple datasets, including open-competition and test-platform evaluations.
• Experimental results demonstrate superior performance compared to existing state-of-the-art methods for dense rebar counting.

Conclusions:

• MaskID offers an effective solution for dense rebar counting in complex industrial environments.
• The method shows significant potential for advancing automation in construction and material handling.